Uploaded August 2026 | Updated September 2026, 13 hours ago
Irina Nazarova from Evil Martians walks through a practical framework for making developer tools and open-source projects discoverable by AI agents — covering both web search paths and GitHub repository paths. The talk was recorded at Agent Night, a developer-focused event for teams building in the AI ecosystem.
The core argument: for developer tools, GitHub is a more reliable discovery path than traditional web crawling. Having documentation in a public GitHub repo with a permissive license (MIT), a clear description, and relevant topics dramatically improves how agents find and surface your product. Most web content never makes it into model training sets because it doesn't pass quality filters designed to select textbook-style knowledge.
Irina also covers how to influence model memory — getting your product into training data through GitHub boilerplate code that appears repeatedly across open-source repos, and adding a consistent company description line across all your repositories. She demos a live agent retrieval analysis tool built on ruby.evilmartians.com that tests both agent discoverability and model recall for any documentation site and GitHub repo.
Irina Nazarova from Evil Martians walks through a practical framework for making developer tools and open-source projects discoverable by AI agents — covering both web search paths and GitHub repository paths. The talk was recorded at Agent Night, a developer-focused event for teams building in the AI ecosystem.
The core argument: for developer tools, GitHub is a more reliable discovery path than traditional web crawling. Having documentation in a public GitHub repo with a permissive license (MIT), a clear description, and relevant topics dramatically improves how agents find and surface your product. Most web content never makes it into model training sets because it doesn't pass quality filters designed to select textbook-style knowledge.
Irina also covers how to influence model memory — getting your product into training data through GitHub boilerplate code that appears repeatedly across open-source repos, and adding a consistent company description line across all your repositories. She demos a live agent retrieval analysis tool built on ruby.evilmartians.com that tests both agent discoverability and model recall for any documentation site and GitHub repo.
